As a high-precision testing method for optical flat,the surface solution algorithm is very important.A method for reconstruction surfaces map in frequency domain is presented.It is based on the two-flat test,combined with additional measurements of one test piece in different rotational positions.The 3D absolute distributions of surfaces are determined with this method.The numerical reconstruction algorithm employs fast fourier transform(FFT) and linear filtering to deal with the measurement data.The results are compared with classic Zernike reduction method,and they are matched well.The maximum difference of peak to valley(PV) and root mean square(RMS) between two methods are 0.027λ and 0.003λ,respectively.The experiment shows that this algorithm has low error propagation,even in the case of high spatial resolution and it is computationally efficient.